Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI – Page 2

Genesis chip may help AI with its memory problem

One of artificial intelligence’s most stubborn problems is enabling AI systems to accumulate new knowledge without losing what they previously learned. A team of researchers at the MATRIX AI Consortium at The University of Texas at San Antonio may have solved this issue with Genesis, a spiking neuromorphic accelerator chip that would enable on-device continual learning throughout its operational lifetime.

Imagine a security drone trained to patrol a dense forest to spot signs of wildfire. After months of honing its ability to identify smoke among pine trees, the drone is reassigned to a coastal region to watch for floods. The moment the drone learns to interpret these new types of images, it might completely lose its ability to detect a forest fire. In the world of artificial intelligence, this phenomenon is known as “catastrophic forgetting,” and it remains one of the biggest hurdles to creating truly intelligent, autonomous agents.

Generative Bionics’ smart robot skin prevents collisions

Humanoids and humans are increasingly sharing factory floors, but one is made of metal and the other of flesh and bones, a mismatch that rarely ends well for us in a collision. Italian startup Generative Bionics thinks it has a fix: a humanoid robot covered in sensing skin that feels people approaching and adjusts its movements before any contact occurs.

The robot, called Gene.01, is wrapped in a network of sensors running from its torso to its limbs. This smart skin tracks touch, temperature, proximity, and force simultaneously, letting the robot anticipate a person’s presence and react before and during contact. It’s a bit like pulling your hand away from a hot stove before you actually touch it. The heat you feel from a distance is enough to make you stop.

The same sensors let humans physically teach the robot new tasks. Rather than only showing Gene.01 a movement on video, a person can guide its arm directly, helping it learn exactly how much force to apply. Generative Bionics says this solves one of humanoid robotics’ persistent headaches: teaching a machine to grip an object firmly enough that it doesn’t drop it but gently enough that it doesn’t crush it – a skill that’s hard to learn from video alone.

Structure and evolutionguided design of minimal RNAguided nucleases

The design of RNA-guided nucleases with properties not limited by evolution can expand programmable genome-editing capabilities. However, generating diverse multidomain proteins with robust enzymatic properties remains challenging. Here, we use a protein design strategy that couples a structure-guided inverse-folding model with evolution-informed residue constraints to generate active, divergent variants of TnpB, a minimal CRISPR-Cas12–like nuclease, termed SynTnpBs. High-throughput screening of artificial intelligence–generated variants yielded editors that retained or exceeded wild-type activity in bacterial, plant, and human cells. Cryo–electron microscopy–based structure determination of the most divergent variant revealed stabilizing contacts in the RNA–DNA interfaces across conformations, demonstrating the design potential of this approach.

Inspired by how children learn, new AI framework learns to theorize the world from observations

A KAIST research team has developed a next-generation world model, an internal model an AI builds to understand and predict the world, that learns executable theories from observation alone.

The team led by Professor Sungjin Ahn from the School of Computing proposed a new learning paradigm called Learning-to-Theorize (L2T), which trains AI to theorize how the world works using only observed information. The team also built the Neural Theorizer (NEO), a neural network-based model that implements this paradigm.

The research was presented at the 43rd International Conference on Machine Learning (ICML 2026), held in Seoul from July 6–11. The paper, published on the arXiv preprint server, was also selected for the Best Paper Award at the Compositional Learning Workshop.

James Martin: We Can Control Accelerating Technology

In February 2011, I spent an hour on Skype asking one of the most influential computer scientists alive whether we could still steer the technologies we were building.

James Martin said yes.

He had earned the right to that answer. Computerworld ranked him fourth among the 25 people who most shaped computer science. The Sunday Times called him Britain’s leading futurist. He wrote 104 textbooks, picked up a Pulitzer nomination, collected honorary doctorates from six continents, then gave away more than $100 million to found the Oxford Martin School so 30 institutes could work on the hardest problems of the century.

So when he told me accelerating technology is controllable, he was not being naive. He was being deliberate. Control, in his telling, was never a technical property of the machines. It was a civilizational choice, and he thought this century was the narrow window in which we get to make it.

We talked about exponential growth in genetics, robotics, nanotech and #AI. We talked about The Meaning of the 21st Century and the project he was working on then, the Transformation of Humankind. He was not selling optimism. He was assigning homework.

Fifteen years later, the claim in the title is a lot harder to defend than it was when he made it. Or maybe that is precisely his point, and we are the ones who failed the assignment.

Continue reading “James Martin: We Can Control Accelerating Technology” | >
/* */